In a visit to HP Labs, we saw a novel approach to displaying and gaining insight from large, multidimensional datasets. The work, called Pixel Bar Charts, builds on perhaps the simplest graphical display, the humble bar chart. Bar charts usually mean that data has been aggregated, often to the point that they provide scant detail. Pixel bar charts build cleverly on this simple bar chart idea. Each bar is constructed of pixels that represent each observation, their position in the bar sorted by a chosen attribute . Color and cartesian bar positon are used to represent other dimensions. Works on continuous and categorical data. You can also click-through on each bar to find individual attributes of observations. The picture above does not do this true justice.
Some interaction similarities to Spotfire, but is more intuitive. My only disappointment is that this is not yet commercially available, I could see this as a big plus for analyzing data, and providing more insight to the casual user of visualization. HP should license it out, appears to have been in existance since about 2001. Here is an HP Labs report on the approach. Here is a real-life example of E-Customer data mining using the approach.
Comments